Building an AI Native Engineering Org - Angie Jones
B · Useful, limited First-hand account from Block's former VP of engineering AI tools and enablement with a concrete maturity model and programme design; the impact figures are internal and self-reported.
Summary based on the full transcript·Original: about 23 minutes to watch
In short
Angie Jones (IT Revolution talk, June 2026) describes moving Block's 3,500-person engineering organisation toward agent-driven work. High tool usage did not speed up delivery, so she built a five-stage maturity model and a hand-picked AI Champions programme, made repositories agent-ready, and enabled delegation from Slack, Jira, Linear and GitHub. The talk ends with layoffs at Block and her question about where this is heading.
Key numbers
- 90%share of Block engineers using coding agents after a few months, while features did not ship faster
- 50hand-picked AI Champions asked to give 30% of their time to team enablement
- 69%rise in AI-authored code about three months after the Champions programme launched
- 21xincrease in automated PRs over the same period; reported time savings rose 37%
- 25,000repositories mapped into a machine-readable company world model for agent orchestration
As reported by the source.
English·Video·opens youtube.com
What it reports
- Maturity stages: 0 no AI; 1 autocomplete; 2 chat without PRs; 3 delegating and reviewing; 4 parallel agents; 5 delegating complete tasks with shippable results. Most engineers sat at stages 1-2 in mid-2025.
- Used the 1-9-90 participation rule: focus on the 1% power users rather than training everyone.
- 50 Champions from critical repositories, each giving 30% of time; not volunteers.
- Repo readiness kit: context files (AGENTS.md, CLAUDE.md), rules files, repeatable workflows or skills, an AI code reviewer with team-specific instructions, AI attribution on PRs, and a readiness score.
- After about three months: AI-authored code up 69%, reported time savings up 37%, automated PRs up 21x. One team cleared a three-sprint workload in the first week.
- Stage 4 bottlenecks: code review overload (handled with Codex review plus an auto-fix agent), agents colliding and laptops overloaded (moved to cloud workstations).
- An internal orchestrator and a world model of 25,000 repositories let agents plan across products; non-engineers could request fixes via Slack.
How they got it
Practitioner talk describing her own programme; no external evaluation.
Limits
- All figures are internal and unaudited; "time savings" is self-reported.
- Productivity metrics are activity counts (code, PRs), not delivery or quality outcomes.
- Block-specific context: its own agent (Goose), monorepos, strong engineering culture.
Why it matters for companies
The sequence is reusable: usage numbers are not impact, pick power users per critical codebase, make the shared asset (the repository) agent-ready so the whole team benefits, then move delegation into the tools where work is assigned. The review bottleneck and the layoff ending are the costs other firms should plan for.
Summary based on the full transcript. Read the original for the full argument and data.Watch the original on youtube.com
- Tags
- AI enablement, Lecture